Research map / Learning theory and algorithms / Prediction and decision losses
Efficient Loss-Based Decoding on Graphs for Extreme Classification
Itay Evron, Edward Moroshko, Koby Crammer
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Increase trellis slice width (ablation) | W-LTLS · sector | test accuracy (%): 88.45 -> 94.88 | Improved |
| Increase trellis slice width (ablation) | W-LTLS · aloi.bin | test accuracy (%): 82.24 -> 95.16 | Improved |
| Increase trellis slice width (ablation) | W-LTLS · imageNet | test accuracy (%): 0.75 -> 12.60 | Improved |
| Increase trellis slice width (ablation) | W-LTLS · Dmoz | test accuracy (%): 23.04 -> 38.89 | Improved |
| Increase trellis slice width (ablation) | W-LTLS · LSHTC1 | test accuracy (%): 9.50 -> 22.47 | Improved |
| Use exponential loss for loss-based decoding instead of squared loss (ablation) | W-LTLS · five extreme-classification datasets | test accuracy | Improved |
| Increase slice width, making induced binary subproblems less balanced (ablation) | W-LTLS · all five datasets | average binary training loss | Improved |
| Prune weights in [-lambda, lambda] (ablation) | W-LTLS · sparse datasets excluding imageNet | test accuracy | No clear effect |
| Use wider graphs before post-pruning (ablation) | W-LTLS · sparse datasets excluding imageNet | model size | Improved |
Closest papers by meaning
- Loss Decomposition for Fast Learning in Large Output Spaces (ICML 2018 · similarity 0.85)
- Breaking the Glass Ceiling for Embedding-Based Classifiers for Large Output Spaces (NeurIPS 2019 · similarity 0.85)
- Label Efficient Learning by Exploiting Multi-Class Output Codes (AAAI 2017 · similarity 0.84)
- PD-Sparse : A Primal and Dual Sparse Approach to Extreme Multiclass and Multilabel Classification (ICML 2016 · similarity 0.84)
- Scalable design of Error-Correcting Output Codes using Discrete Optimization with Graph Coloring (NeurIPS 2022 · similarity 0.84)
- Taming the Sigmoid Bottleneck: Provably Argmaxable Sparse Multi-Label Classification (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
